Re: [PATCH v3] EXPR: Emit an truncate if 31+ bits polluted for SImode

[Jeff Law <jeffreyalaw@gmail.com>]

On 1/3/24 16:39, Richard Sandiford wrote:
> YunQiang Su <syq@gcc.gnu.org> writes:
>> On TRULY_NOOP_TRUNCATION_MODES_P (DImode, SImode)) == true platforms,
>> if 31 or above bits is polluted by an bitops, we will need an
>> truncate. Let's emit one, and mark let's use the same hardreg
>> as in and out, the RTL may like:
>>
>> (insn 21 20 24 2 (set (subreg/s/u:SI (reg/v:DI 200 [ val ]) 0)
>>          (truncate:SI (reg/v:DI 200 [ val ]))) "../xx.c":7:29 -1
>>       (nil))
>>
>> We use /s/u flags to mark it as really needed, as in
>> combine_simplify_rtx, this insn may be considered as truncated,
>> so let's skip this combination.
>>
>> gcc/ChangeLog:
>>          PR: 104914.
>>          * combine.cc (try_combine): Skip combine with truncate if
>> 	dest is subreg and has /u/s flags on platforms
>> 	TRULY_NOOP_TRUNCATION_MODES_P (DImode, SImode)) == true.
>> 	* expr.cc (expand_assignment): Emit a truncate insn, if
>> 	31+ bits is polluted for SImode.
>>
>> gcc/testsuite/ChangeLog:
>> 	PR: 104914.
>> 	* gcc.target/mips/pr104914.c: New testcase.
> 
> Sorry for not looking at this earlier.  I've got a bit lost in the
> various threads, so apologies if this has been discussed already
> but I think the fix is:
> 
> diff --git a/gcc/expr.cc b/gcc/expr.cc
> index 4f42c0ff487..9847eba19fe 100644
> --- a/gcc/expr.cc
> +++ b/gcc/expr.cc
> @@ -6275,9 +6275,8 @@ expand_assignment (tree to, tree from, bool nontemporal)
>   	      else
>   		{
>   		  rtx to_rtx1
> -		    = lowpart_subreg (subreg_unpromoted_mode (to_rtx),
> -				      SUBREG_REG (to_rtx),
> -				      subreg_promoted_mode (to_rtx));
> +		    = convert_to_mode (subreg_unpromoted_mode (to_rtx),
> +				       SUBREG_REG (to_rtx), false);
>   		  result = store_field (to_rtx1, bitsize, bitpos,
>   					bitregion_start, bitregion_end,
>   					mode1, from, get_alias_set (to),
> 
> (completely untested apart from the test case).  That should still
> produce a subreg on most targets, but generates the required trunc
> on MIPS.
I think this is a subset of Roger's work that got committed a few days back.

jeff